Pith. sign in

REVIEW 3 major objections 5 minor 61 references

A single scalar from the reference, plus Hermite-Gauss self-prediction of the distorted gradient, is enough for PreSPA to match leading no-reference image-quality scores and approach full-reference accuracy on structural distortions.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

T0 review · grok-4.5

2026-07-13 06:34 UTC pith:OEREHYOI

load-bearing objection Solid PR-IQA engineering that really does reduce the reference to one scalar and competes with deep NR on structural distortions; the 'no dataset-specific calibration' phrasing is oversold because (b0,b1,b2) are still fit per dataset. the 3 major comments →

arxiv 2607.08563 v2 pith:OEREHYOI submitted 2026-07-09 eess.IV

Partial-Reference IQA Based on Hermite-Gauss Structural Prediction and Texture Deviation

Paolo Giannitrapani , Elio D. Di Claudio , Giovanni Jacovitti This is my paper
classification eess.IV
keywords partial-reference IQAHermite-Gauss structural predictiontexture deviationviewing-distance modelingtraining-free quality metriccomplex gradient fieldedge-texture leakage
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

Image quality assessment usually demands either the full pristine image or a large trained network. PreSPA claims neither is necessary. It extracts one scalar prior μ from energy differences between reference and distorted complex gradients on strong edges, then spreads that prior over weakly structured regions to capture how edge damage leaks into surrounding texture. A fully no-reference structural index is obtained by regressing the distorted gradient field onto second-order Hermite-Gauss bases and reading the residual divergence and orientation variance as a self-channel capacity. The two indices share a viewing-distance-dependent scale and fuse with three affine coefficients, without dataset-specific logistic calibration. Across six public benchmarks the method rivals or exceeds deep no-reference models and, on structural distortions, approaches traditional full-reference metrics while using only 0.13 GFLOPS and three parameters.

Core claim

PreSPA shows that a single scalar prior μ, computed once from localized gradient-energy deviations on strong-edge regions and accumulated on weak-structure regions, is sufficient to drive a texture index and to modulate a Hermite-Gauss self-prediction structural index so that their three-parameter affine fusion matches or exceeds leading no-reference quality predictors and approaches full-reference accuracy on structural distortions, all without learned weights or post-hoc VQEG rectification.

What carries the argument

The scalar prior μ together with Hermite-Gauss self-prediction of the distorted complex gradient: second-order bases yield a coefficient field G(p) whose divergence and angular variance form a reliability weight that modulates μ into a self-channel capacity Q_struct; μ itself is the sole reference footprint.

Load-bearing premise

Natural-image complex gradients remain self-predictable enough under second-order Hermite-Gauss regression that residual divergence and orientation variance, scaled only by the single scalar μ, track perceived structural damage without further reference information.

What would settle it

On a controlled set of common structural distortions (blur, noise, compression) for which full-reference structural metrics remain high, measure whether the Hermite-Gauss residual and resulting Q_struct stay low; if they rise while human scores stay high, the self-prediction premise fails.

Watch this falsifier — get emailed when new claim-graph text bears on it.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

3 major / 5 minor

Summary. PreSPA is a Partial-Reference IQA method that reduces the reference to a single scalar prior μ, computed once from complex-gradient energy differences on strong-edge regions (Ω_C) and accumulated over weakly structured regions (Ω_H). A fully No-Reference structural index Q_struct is obtained by second-order Hermite-Gauss self-prediction of the distorted gradient field, using divergence and angular variance of the coefficient field as a reliability-weighted, rate-distortion-style capacity. A texture index Q_t is a saturating SNR driven by μ. Viewing distance enters the Gaussian operator scale; the two indices are fused by a three-coefficient affine map. On six benchmarks the method reports SROCC/PLCC that rival leading deep NR models on structural distortions and approach classical FR metrics on LIVE DBR2, CSIQ and LIVE MD, while remaining weak on colour/contrast classes by design.

Significance. The work addresses an under-explored niche (minimal-footprint PR-IQA) with a clear perceptual story (edge–texture leakage, Marr-style coupling) and a genuinely compact design: one reference scalar, three fusion coefficients, ~0.13 GFLOPS, no learned weights. Embedding viewing distance in the operator scale and replacing VQEG logistic with affine alignment is a useful transfer of the BELE principle to the PR setting. Per-distortion tables on six datasets, including PIPAL, and an honest account of colour failures, make the empirical case falsifiable and useful. If the self-predictability premise holds under the stated operators, the paper offers a transparent, deployable alternative to large NR networks for structural monitoring tasks.

major comments (3)
  1. Abstract and Sec. I claim “no dataset-specific calibration,” yet Sec. IV-D and V-A state that (b0,b1,b2) “are obtained, for each dataset, through the same robust weighted affine regression against the DMOS scale.” Because PreSPA = b0 + b1 a Q_struct + b2 a Q_t, the relative weights b1/b2 can change rankings whenever the two indices disagree; the fit is therefore not a pure monotonic re-scaling and can improve SROCC as well as PLCC. The abstract wording overstates the training-free claim and should be aligned with the body (e.g., “affine alignment of three coefficients in place of five-parameter VQEG logistic”).
  2. No experiment freezes (b0,b1,b2) across datasets or reports leave-one-dataset-out transfer of the fusion weights. Without that, the headline cross-benchmark competitiveness (Tables I–III) is partly confounded by per-dataset re-weighting of Q_struct vs Q_t. A short table with global or transferred coefficients (even if weaker on colour-heavy aggregates) is needed to substantiate the “training-free / calibration-light” narrative that distinguishes PreSPA from deep NR models.
  3. Sec. IV-C: the central NR assumption—that residual divergence and angular variance of the Hermite-Gauss field G(p) track perceived structural degradation when scaled only by μ—is supported mainly by qualitative maps (Figs. 2–3) and end-to-end scores. A minimal ablation (Q_struct alone, Q_t alone, and μ replaced by a constant) on at least LIVE DBR2 and PIPAL would show how much of the reported accuracy is carried by the self-prediction branch versus the reference scalar, and would directly test the weakest modelling assumption.
minor comments (5)
  1. Table I header lists FLOPS/parameters for NR deep methods but places PreSPA in the same block; a clearer separation of PR vs NR columns would avoid implying PreSPA is NR.
  2. Notation: M = √0.3 is introduced as a certainty threshold without a short justification of the −10 dB choice beyond “chosen empirically”; one sentence or a pointer to the supplementary sensitivity analysis would help.
  3. Sec. IV-B: y_err is written as y|Ω_C − ỹ|Ω_C; clarify whether this is a complex subtraction restricted to the mask or a masked residual after full-field computation.
  4. Related work (Sec. II-B) could briefly cite more recent RR/PR or degraded-reference work beyond RRED and free-energy models so the “single scalar” novelty claim is better situated.
  5. Typos / style: “PROPOSAL2026” page headers; occasional missing spaces before citations; “Clip-based” capitalisation in Table III.

Circularity Check

2 steps flagged

Per-dataset affine fit of (b0,b1,b2) to DMOS undercuts the abstract's 'no dataset-specific calibration' claim; reported correlations partially rely on this alignment rather than pure operator predictions.

specific steps
  1. fitted input called prediction [Abstract; Sec. IV-D Eq. (11); Sec. V-A]
    "The final score is produced by an affine fusion with only three interpretable parameters, making the method compact, transparent, and computationally efficient, with the viewing distance embedded into the operator scale and no dataset-specific calibration. ... They are obtained, for each dataset, through the same robust weighted affine regression against the DMOS scale adopted in [12]; since the combination is affine and the indices are already perceptually aligned, this fitting is a per-dataset scale-and-offset alignment, not a data-driven training stage."

    Abstract and contributions assert 'no dataset-specific calibration' and training-free operation with three parameters. Yet the PreSPA score (and thus all SROCC/PLCC in Tables I-III supporting the claim of rivalling NR deep models) is produced only after fitting those three coefficients per dataset to DMOS. The relative weighting of Qstruct versus Qt is therefore dataset-dependent; the reported correlations are not pure predictions from the Hermite-Gauss and mu operators alone but include a forced scale-and-offset alignment that improves match by construction.

  2. self citation load bearing [Sec. I contributions; Sec. III; Sec. IV-D]
    "PreSPA is, to our knowledge, the first PR method to adopt this principle, recently introduced for the FR estimator BELE. ... We briefly recall the Virtual Receptive Field model introduced in [12], [11], as a foundation for defining the concept of visual map ... The two a priori parameters (a, τ) mirror the role of the pair (Q, τ) in the Full-Reference estimator BELE [12], [11]"

    The key claimed novelty that eliminates VQEG rectification—embedding viewing distance directly into operator scale so that Qt and Qstruct already share a common perceptual scale—is imported wholesale from the authors' own prior FR papers BELE without re-derivation or external validation here. The PR contribution is presented as the first application of that self-cited principle; the calibration-free narrative therefore rests on the prior self-work rather than an independent argument.

full rationale

The core indices Qt and Qstruct are defined from image operators (complex gradients, Hermite-Gauss regression on the distorted field alone, energy residuals for mu) without reference to subjective scores, so the structural/texture decomposition itself is not circular. Viewing-distance scaling of kernels is imported from the authors' prior BELE work but is applied as a fixed physical rescaling, not a uniqueness theorem forbidding alternatives. The circularity is limited to the final stage: the paper repeatedly claims a training-free estimator with 'no dataset-specific calibration' that reduces VQEG logistic rectification to plain affine alignment, yet Sec. IV-D and V-A explicitly obtain the three fusion coefficients by robust weighted regression against each dataset's DMOS. Because the relative weights b1/b2 affect the linear combination, both PLCC and (to a lesser extent) SROCC benefit from this per-benchmark fit; the headline numbers that 'rival or exceed' deep NR models therefore include a dataset-specific alignment step that the abstract and contributions list simultaneously deny. This is standard IQA practice and does not collapse the operator definitions, but it makes the strongest performance claims partially fitted rather than pure first-principles predictions, warranting a moderate score of 4 rather than 0 or 6+.

Axiom & Free-Parameter Ledger

7 free parameters · 5 axioms · 3 invented entities

The central claim rests on the authors' prior Virtual Receptive Field / BELE framework, Marr-style edge-to-texture leakage, a self-channel capacity interpretation of structural integrity, several hand-chosen thresholds and rates, and three per-dataset affine coefficients. No new physical entities are postulated; the invented constructs are algorithmic (mu, Omega_C/H, G(p), Q_struct).

free parameters (7)
  • b0, b1, b2 (affine fusion coefficients)
    Fitted per dataset by robust weighted regression to DMOS; the only free parameters counted by the authors, yet still data-dependent scale-and-offset alignment.
  • a (statistical quality anchor)
    Fixed a priori from metadata or blur regression; scales the scoring system to each database lower bound.
  • tau = delta/delta_0 (normalized viewing distance)
    From metadata or recovered by regression on Gaussian-blurred images; rescales all Gaussian kernels as sigma = sigma_0 * tau^2.
  • M = sqrt(0.3) (certainty threshold)
    Empirical -10 dB drop used to split Omega_C vs Omega_H; chosen by hand.
  • C = 20 (texture noise floor)
    Constant in Q_t = C/(mu+C); sets the saturating SNR floor.
  • gamma_0 = 10, gamma_1 = 1 (curvature weights)
    Balance divergence vs angular variance and set exponential reliability rate; fixed across datasets with only a sensitivity note in supplementary.
  • sigma_w = 10 (Hermite-Gauss regression window)
    Wide Gaussian support for local regression of B(p); fixed by design.
axioms (5)
  • domain assumption Virtual Receptive Field model and viewing-distance scaling of retinal bandwidth (Sec. III, inherited from BELE).
    All gradient maps and operator scales rest on this retinal/display model; not re-derived here.
  • domain assumption Marr raw-primal-sketch hierarchy: strong-edge degradations perceptually leak into surrounding texture (Sec. IV-B).
    Justifies measuring error on Omega_C but accumulating on Omega_H and normalizing by |Omega_C|.
  • ad hoc to paper Natural-image complex gradients are locally self-predictable by second-order Hermite-Gauss bases so residual tracks structural degradation (Sec. IV-C).
    Core modeling choice enabling the NR structural index; not independently proven.
  • domain assumption Quality can be framed as normalized self-channel capacity under additive noise modulated by structural reliability (Q_struct formula).
    Information-theoretic interpretation borrowed from VIF-style thinking but applied without a reference.
  • ad hoc to paper Affine fusion of already scale-aligned indices replaces five-parameter VQEG logistic without loss of validity.
    Stated as a design principle; validity is empirical per dataset.
invented entities (3)
  • scalar prior mu (cross-regional edge-to-texture energy deviation) no independent evidence
    purpose: Sole reference footprint; modulates both texture SNR and structural noise variance.
    Defined by the paper's dual-region construction; no external independent measurement of 'perceptual leakage' is supplied.
  • Hermite-Gauss structural field G(p) and reliability weight w(p) no independent evidence
    purpose: Reference-free descriptor of curvature integrity and orientation coherence.
    Constructed from distorted image only via weighted regression onto phi_20/phi_02; purpose-built for this estimator.
  • certainty / weaker-edge partition Omega_C / Omega_H no independent evidence
    purpose: Spatial mask that isolates where mu is measured versus where it is accumulated.
    Algorithmic partition controlled by hand-chosen M; not an independently observed anatomical or psychophysical region.

pith-pipeline@v1.1.0-grok45 · 32045 in / 3840 out tokens · 38437 ms · 2026-07-13T06:34:52.341297+00:00 · methodology

0 comments
read the original abstract

We propose PreSPA (Partial-Reference Structural Prediction Approach), a Partial-Reference Image Quality Assessment framework that decomposes perceptual quality into two complementary indices. A structure-aware index, operating in a No-Reference manner, captures structural degradation through Hermite-Gauss prediction of the distorted gradient field and the angular variance of its curvature. A texture-sensitive index estimates local noise through a scalar prior $\mu$, obtained from energy differences between reference and distorted complex gradient maps on strong-edge regions and accumulated over weakly-structured ones, reflecting the perceptual leakage of degraded edges into surrounding textures. Crucially, $\mu$ is the only information extracted from the reference and is computed once per image pair, reducing the reference footprint to a single scalar value. The final score is produced by an affine fusion with only three interpretable parameters, making the method compact, transparent, and computationally efficient, with the viewing distance embedded into the operator scale and no dataset-specific calibration. Extensive evaluations on six standard benchmarks show that PreSPA consistently rivals or exceeds leading No-Reference approaches, while in several cases matching the accuracy of Full-Reference models.

Figures

Figures reproduced from arXiv: 2607.08563 by Elio D. Di Claudio, Giovanni Jacovitti, Paolo Giannitrapani.

Figure 1
Figure 1. Figure 1: Flowchart illustrating the extended computational pipeline for deriving the texture index [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Upper row: Gaussian blurred images ”Sailing1” from the LIVE DBR2 dataset, evaluated at Gaussian blur levels of [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Upper row: Gaussian blurred images ”Churchandcapitol” from the LIVE DBR2 dataset, evaluated at Gaussian blur levels of [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗

Review history (2 revisions) →

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

61 extracted references · 2 linked inside Pith

  1. [1]

    Image quality assessment: from error visibility to structural similarity,

    Z. Wang, A. C. Bovik, H. R. Sheikh, and E. P. Simoncelli, “Image quality assessment: from error visibility to structural similarity,”IEEE Trans. on Image Processing, vol. 13, no. 4, pp. 600–612, Apr. 2004

    2004

  2. [2]

    A statistical evaluation of recent Full Reference image quality assessment algorithms,

    H. R. Sheikh, M. F. Sabir, and A. C. Bovik, “A statistical evaluation of recent Full Reference image quality assessment algorithms,”IEEE Trans. on Image Processing, vol. 15, no. 11, pp. 3441–3452, Nov. 2006. [Online]. Available: live.ece.utexas.edu/research/quality/subjective.htm

    2006

  3. [3]

    Review of medical image quality assessment,

    L. S. Chow and R. Paramesran, “Review of medical image quality assessment,”Biomedical Signal Processing and Control, vol. 27, pp. 145–154, 2016. [Online]. Available: https://www.sciencedirect.com/ science/article/pii/S1746809416300180

    2016

  4. [4]

    Blindly assess image quality in the wild guided by a self-adaptive hyper network,

    S. Su, Q. Yan, Y . Zhu, C. Zhang, X. Ge, J. Sun, and Y . Zhang, “Blindly assess image quality in the wild guided by a self-adaptive hyper network,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), June 2020

    2020

  5. [5]

    Musiq: Multi- scale image quality transformer,

    J. Ke, Q. Wang, Y . Wang, P. Milanfar, and F. Yang, “Musiq: Multi- scale image quality transformer,” inProceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), October 2021, pp. 5148–5157

    2021

  6. [6]

    Topiq: A top-down approach from semantics to distortions for image quality assessment,

    C. Chen, J. Mo, J. Hou, H. Wu, L. Liao, W. Sun, Q. Yan, and W. Lin, “Topiq: A top-down approach from semantics to distortions for image quality assessment,”IEEE Transactions on Image Processing, vol. 33, pp. 2404–2418, 2024

    2024

  7. [7]

    Exploring clip for assessing the look and feel of images,

    J. Wang, K. C. Chan, and C. C. Loy, “Exploring clip for assessing the look and feel of images,”Proceedings of the AAAI Conference on Artificial Intelligence, vol. 37, no. 2, pp. 2555–2563, Jun. 2023. [Online]. Available: https://ojs.aaai.org/index.php/AAAI/article/view/25353

    2023

  8. [8]

    Reduced-reference image quality as- sessment using a wavelet-domain natural image statistic model,

    Z. Wang and E. P. Simoncelli, “Reduced-reference image quality as- sessment using a wavelet-domain natural image statistic model,”Proc. of SPIE Human Vision and Electronic Imaging X, vol. 5666, pp. 149– 159, Mar. 2005

    2005

  9. [9]

    RRED indices: Reduced reference entropic differencing for image quality assessment,

    R. Soundararajan and A. C. Bovik, “RRED indices: Reduced reference entropic differencing for image quality assessment,”IEEE Transactions on Image Processing, vol. 21, no. 2, pp. 517–526, Feb. 2012

    2012

  10. [10]

    Marr,Vision: A Computational Investigation into the Human Repre- sentation and Processing of Visual Information

    D. Marr,Vision: A Computational Investigation into the Human Repre- sentation and Processing of Visual Information. The MIT Press, 2010

    2010

  11. [11]

    Bele: Blur equivalent linearized estimator,

    P. Giannitrapani, E. D. D. Claudio, and G. Jacovitti, “Bele: Blur equivalent linearized estimator,” 2025. [Online]. Available: https://arxiv.org/abs/2503.00503

    Pith/arXiv arXiv 2025

  12. [12]

    Full-reference calibration-free image quality assessment,

    P. Giannitrapani, E. D. Di Claudio, and G. Jacovitti, “Full-reference calibration-free image quality assessment,”Signal Processing: Image Communication, vol. 130, p. 117212, 2025. [Online]. Available: https://www.sciencedirect.com/science/article/pii/S0923596524001139

    2025

  13. [13]

    Final report from the video quality experts group on the validation of objective models of video quality assessment,

    VQEG, “Final report from the video quality experts group on the validation of objective models of video quality assessment,” Video Quality Experts Group, http://www.vqeg.org, Tech. Rep., Mar. 2000

    2000

  14. [14]

    Issues in vision modeling for perceptual video quality assessment,

    S. Winkler, “Issues in vision modeling for perceptual video quality assessment,”Signal Processing, vol. 78, no. 2, pp. 231–252, Feb. 1999

    1999

  15. [15]

    VSNR: a wavelet-based visual signal- to-noise ratio for natural images,

    D. M. Chandler and S. Hemami, “VSNR: a wavelet-based visual signal- to-noise ratio for natural images,”IEEE Trans. on Image Processing, vol. 16, no. 9, pp. 2284 –2298, Sept. 2007

    2007

  16. [16]

    Effects of spatial correlations and global precedence on the visual fidelity of distorted images,

    D. M. Chandler, K. H. Lim, and S. S. Hemami, “Effects of spatial correlations and global precedence on the visual fidelity of distorted images,” inProc. SPIE, vol. 6057, 2006, pp. 60 570F–60 570F–15. [Online]. Available: http://dx.doi.org/10.1117/12.655442

    doi:10.1117/12.655442 2006

  17. [17]

    Visual fixation patterns when judging image quality: Effects of distortion type, amount, and subject experience,

    E. Vu and D. Chandler, “Visual fixation patterns when judging image quality: Effects of distortion type, amount, and subject experience,” in Image Analysis and Interpretation, 2008. SSIAI 2008. IEEE Southwest Symposium on, Mar. 2008, pp. 73–76

    2008

  18. [18]

    Visual cognition,

    P. Cavanagh, “Visual cognition,”Vision Research, vol. 51, no. 13, pp. 1538 – 1551, 2011, vision Research 50th Anniversary Issue: Part 2. PROPOSAL2026 13

    2011

  19. [19]

    FSIM: A feature similarity index for image quality assessment,

    L. Zhang, D. Zhang, M. Xuanqin, and D. Zhang, “FSIM: A feature similarity index for image quality assessment,”IEEE Trans. on Image Processing, vol. 20, no. 8, pp. 2378–2386, Aug. 2011

    2011

  20. [20]

    Gradient magnitude similarity deviation: A highly efficient perceptual image quality index,

    W. Xue, L. Zhang, X. Mou, and A. Bovik, “Gradient magnitude similarity deviation: A highly efficient perceptual image quality index,” IEEE Trans. on Image Processing, vol. 23, no. 2, pp. 684–695, Feb. 2014

    2014

  21. [21]

    Most apparent distortion: full- reference image quality assessment and the role of strategy,

    E. C. Larson and D. M. Chandler, “Most apparent distortion: full- reference image quality assessment and the role of strategy,”J. Electron. Imaging, vol. 9, no. 1, pp. 011 006/1–21, Jan. 2010

    2010

  22. [22]

    Image information and visual quality,

    H. R. Sheikh and A. C. Bovik, “Image information and visual quality,” IEEE Trans. on Image Processing, vol. 15, no. 2, pp. 430–444, Feb. 2006

    2006

  23. [23]

    Multi-scale structural similarity for image quality assessment,

    Z. Wang, E. Simoncelli, and A. Bovik, “Multi-scale structural similarity for image quality assessment,” inProc. of the 37th IEEE Asilomar Conference on Signal, Systems and Computer, vol. 2. IEEE Comp. Soc., Nov. 2003, pp. 1398–1402

    2003

  24. [24]

    Predicting the quality of images compressed after distortion in two steps,

    X. Yu, C. G. Bampis, P. Gupta, and A. C. Bovik, “Predicting the quality of images compressed after distortion in two steps,”IEEE Transactions on Image Processing, vol. 28, no. 12, pp. 5757–5770, 2019

    2019

  25. [25]

    Rvsim: a feature similarity method for full-reference image quality assessment,

    G. Yang, D. Li, F. Lu, Y . Liao, and W. Yang, “Rvsim: a feature similarity method for full-reference image quality assessment,”EURASIP Journal on Image and Video Processing, vol. 2018, no. 1, p. 6, Jan 2018. [Online]. Available: https://doi.org/10.1186/s13640-018-0246-1

    doi:10.1186/s13640-018-0246-1 2018

  26. [26]

    Saliency-guided local full-reference image quality assess- ment,

    D. Varga, “Saliency-guided local full-reference image quality assess- ment,”Signals, vol. 3, no. 3, pp. 483–496, 2022

    2022

  27. [27]

    Full-reference image quality assessment based on gr ¨unwald– letnikov derivative, image gradients, and visual saliency,

    ——, “Full-reference image quality assessment based on gr ¨unwald– letnikov derivative, image gradients, and visual saliency,”Electronics, vol. 11, no. 4, p. 559, 2022

    2022

  28. [28]

    Pipal: A large-scale image quality assessment dataset for perceptual image restoration,

    G. Jinjin, C. Haoming, C. Haoyu, Y . Xiaoxing, J. S. Ren, and D. Chao, “Pipal: A large-scale image quality assessment dataset for perceptual image restoration,” inComputer Vision – ECCV 2020, A. Vedaldi, H. Bischof, T. Brox, and J.-M. Frahm, Eds. Cham: Springer Inter- national Publishing, 2020, pp. 633–651

    2020

  29. [29]

    Image quality assess- ment: Unifying structure and texture similarity,

    K. Ding, K. Ma, S. Wang, and E. P. Simoncelli, “Image quality assess- ment: Unifying structure and texture similarity,”IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 44, no. 5, pp. 2567– 2581, 2022

    2022

  30. [30]

    The unreasonable effectiveness of deep features as a perceptual metric,

    R. Zhang, P. Isola, A. A. Efros, E. Shechtman, and O. Wang, “The unreasonable effectiveness of deep features as a perceptual metric,” in Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), June 2018

    2018

  31. [31]

    Pieapp: Perceptual image- error assessment through pairwise preference,

    E. Prashnani, H. Cai, Y . Mostofi, and P. Sen, “Pieapp: Perceptual image- error assessment through pairwise preference,” inProceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), June 2018

    2018

  32. [32]

    Sg-jnd: Semantic-guided just noticeable distortion predictor for image compression,

    L. Cao, W. Sun, X. Min, J. Jia, Z. Zhang, Z. Chen, Y . Zhu, L. Liu, Q. Chen, J. Chen, and G. Zhai, “Sg-jnd: Semantic-guided just noticeable distortion predictor for image compression,” in2024 IEEE International Conference on Image Processing (ICIP), 2024, pp. 1139–1145

    2024

  33. [33]

    Perceptual image quality assessment with transformers,

    M. Cheon, S.-J. Yoon, B. Kang, and J. Lee, “Perceptual image quality assessment with transformers,” inProceedings of the IEEE/CVF Confer- ence on Computer Vision and Pattern Recognition (CVPR) Workshops, June 2021, pp. 433–442

    2021

  34. [34]

    Convolved quality transformer: Image quality assessment via long-range interaction between local perception,

    H. Oh, J. Kim, T. Kim, and S. Lee, “Convolved quality transformer: Image quality assessment via long-range interaction between local perception,”IEEE Access, vol. 10, pp. 102 968–102 980, 2022

    2022

  35. [35]

    Attentions help cnns see better: Attention-based hybrid image quality assessment network,

    S. Lao, Y . Gong, S. Shi, S. Yang, T. Wu, J. Wang, W. Xia, and Y . Yang, “Attentions help cnns see better: Attention-based hybrid image quality assessment network,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops, June 2022, pp. 1140–1149

    2022

  36. [36]

    Q-Align: Teaching LMMs for visual scoring via discrete text-defined levels,

    H. Wu, Z. Zhang, W. Zhang, C. Chen, L. Liao, C. Li, Y . Gao, A. Wang, E. Zhang, W. Sun, Q. Yan, X. Min, G. Zhai, and W. Lin, “Q-Align: Teaching LMMs for visual scoring via discrete text-defined levels,” in Proc. of the 41st Int. Conf. on Machine Learning (ICML), 2024

    2024

  37. [37]

    Adaptive image quality assessment via teaching large multimodal model to compare,

    H. Zhu, X. Sui, B. Chen, X. Liu, P. Chen, Y . Fang, and S. Wang, “Adaptive image quality assessment via teaching large multimodal model to compare,” inAdvances in Neural Information Processing Systems (NeurIPS), 2024

    2024

  38. [38]

    Blind image quality assessment: From natural scene statistics to perceptual quality,

    A. K. Moorthy and A. C. Bovik, “Blind image quality assessment: From natural scene statistics to perceptual quality,” vol. 20, no. 12, Dec. 2011, pp. 3350–3364

    2011

  39. [39]

    Blind image quality assessment: A natural scene statistics approach in the DCT domain,

    M. A. Saad, A. C. Bovik, and C. Charrier, “Blind image quality assessment: A natural scene statistics approach in the DCT domain,” IEEE Transactions on Image Processing, vol. 21, no. 8, pp. 3339–3352, Aug 2012

    2012

  40. [40]

    No-reference image quality assessment in the spatial domain,

    A. Mittal, A. K. Moorthy, and A. C. Bovik, “No-reference image quality assessment in the spatial domain,”IEEE Transactions on Image Processing, vol. 21, no. 12, pp. 4695–4708, Dec 2012

    2012

  41. [41]

    Making a “completely blind

    A. Mittal, R. Soundararajan, and A. C. Bovik, “Making a “completely blind” image quality analyzer,”IEEE Signal Processing Letters, vol. 20, no. 3, pp. 209–212, 2013

    2013

  42. [42]

    Convolutional neural networks for no-reference image quality assessment,

    L. Kang, P. Ye, Y . Li, and D. Doermann, “Convolutional neural networks for no-reference image quality assessment,” inProc. of the IEEE Conf. on Computer Vision and Pattern Recognition (CVPR), 2014, pp. 1733– 1740

    2014

  43. [43]

    MANIQA: Multi-dimension attention network for no-reference image quality assessment,

    S. Yang, T. Wu, S. Shi, S. Lao, Y . Gong, M. Cao, J. Wang, and Y . Yang, “MANIQA: Multi-dimension attention network for no-reference image quality assessment,” inProc. of the IEEE/CVF Conf. on Computer Vision and Pattern Recognition (CVPR) Workshops, 2022, pp. 1191–1200

    2022

  44. [44]

    Re-IQA: Unsupervised learning for image quality assessment in the wild,

    A. Saha, S. Mishra, and A. C. Bovik, “Re-IQA: Unsupervised learning for image quality assessment in the wild,” inProc. of the IEEE/CVF Conf. on Computer Vision and Pattern Recognition (CVPR), 2023, pp. 5846–5855

    2023

  45. [45]

    ARNIQA: Learning distortion manifold for image quality assessment,

    L. Agnolucci, L. Galteri, M. Bertini, and A. Del Bimbo, “ARNIQA: Learning distortion manifold for image quality assessment,” inProc. of the IEEE/CVF Winter Conf. on Applications of Computer Vision (WACV), 2024, pp. 189–198

    2024

  46. [46]

    A comprehensive performance evaluation of image quality assessment algorithms,

    S. Athar and Z. Wang, “A comprehensive performance evaluation of image quality assessment algorithms,”IEEE Access, vol. 7, pp. 140 030– 140 070, 2019

    2019

  47. [47]

    No-reference image quality assessment metric by combining free energy theory and structural degradation model,

    K. Gu, G. Zhai, X. Yang, W. Zhang, and L. Liang, “No-reference image quality assessment metric by combining free energy theory and structural degradation model,” in2013 IEEE Int. Conf. on Multimedia and Expo (ICME), July 2013, pp. 1–6

    2013

  48. [48]

    Two-dimensional approach to full-reference image quality assessment based on positional structural information,

    L. Capodiferro, G. Jacovitti, and E. D. Di Claudio, “Two-dimensional approach to full-reference image quality assessment based on positional structural information,”IEEE Trans. on Image Processing, vol. 21, no. 2, pp. 505–516, Feb. 2012

    2012

  49. [49]

    Quality assessment considering viewing distance and image resolution,

    K. Gu, M. Liu, G. Zhai, X. Yang, and W. Zhang, “Quality assessment considering viewing distance and image resolution,”IEEE Transactions on Broadcasting, vol. 61, no. 3, pp. 520–531, 2015

    2015

  50. [50]

    Deep neural networks for no-reference and full-reference image quality as- sessment,

    S. Bosse, D. Maniry, K. Muller, T. Wiegand, and W. Samek, “Deep neural networks for no-reference and full-reference image quality as- sessment,”IEEE Transactions on Image Processing, vol. 27, no. 1, pp. 206–219, 2018

    2018

  51. [51]

    Sampling, data transmission, and the Nyquist rate,

    H. Landau, “Sampling, data transmission, and the Nyquist rate,”Pro- ceedings of the IEEE, vol. 55, no. 10, pp. 1701–1706, 1967

    1967

  52. [52]

    Predicting blur vi- sual discomfort for natural scenes by the loss of positional information,

    E. D. Di Claudio, P. Giannitrapani, and G. Jacovitti, “Predicting blur vi- sual discomfort for natural scenes by the loss of positional information,” Vision Research, vol. 189, pp. 33–45, 2021

    2021

  53. [53]

    Maximum Likelihood orientation estimation of 1-D patterns in Laguerre-Gauss subspaces,

    E. D. Di Claudio, G. Jacovitti, and A. Laurenti, “Maximum Likelihood orientation estimation of 1-D patterns in Laguerre-Gauss subspaces,” IEEE Trans. on Image Processing, vol. 19, no. 5, pp. 1113–1125, May 2010

    2010

  54. [54]

    Optical and retinal factors affecting visual resolution,

    F. W. Campbell and D. G. Green, “Optical and retinal factors affecting visual resolution,”The Journal of physiology, vol. 181, no. 3, pp. 576– 93, 1965

    1965

  55. [55]

    The effects of a visual fidelity criterion of the encoding of images,

    J. Mannos and D. Sakrison, “The effects of a visual fidelity criterion of the encoding of images,”IEEE Trans. on Information Theory, vol. 20, no. 4, pp. 525–536, Jul. 1974

    1974

  56. [56]

    Six formal properties of two-dimensional anisotropic visual filters: structural principles and frequency/orientation selectivity,

    J. G. Daugman, “Six formal properties of two-dimensional anisotropic visual filters: structural principles and frequency/orientation selectivity,” IEEE Trans. Syst., Man, Cybern., vol. 13, pp. 882–887, 1983

    1983

  57. [57]

    Maximum likelihood localization of 2-D patterns in the Gauss-Laguerre transform domain: theoretic framework and preliminary results,

    A. Neri and G. Jacovitti, “Maximum likelihood localization of 2-D patterns in the Gauss-Laguerre transform domain: theoretic framework and preliminary results,”IEEE Trans. on Image Processing, vol. 13, no. 1, pp. 72–86, Jan. 2004

    2004

  58. [58]

    Image database TID2013: Peculiarities, results and perspectives,

    N. Ponomarenko, L. Jin, O. Ieremeiev, V . Lukin, K. Egiazarian, J. Astola, B. V ozel, K. Chehdi, M. Carli, F. Battisti, and C.-C. J. Kuo], “Image database TID2013: Peculiarities, results and perspectives,”Signal Pro- cessing: Image Communication, vol. 30, pp. 57 – 77, 2015

    2015

  59. [59]

    Objective quality assessment of multiply distorted images,

    D. Jayaraman, A. Mittal, A. K. Moorthy, and A. C. Bovik, “Objective quality assessment of multiply distorted images,” in2012 Conference Record of the Forty Sixth Asilomar Conference on Signals, Systems and Computers (ASILOMAR), 2012, pp. 1693–1697

    2012

  60. [60]

    Deepfl-iqa: Weak supervision for deep iqa feature learning,

    H. Lin, V . Hosu, and D. Saupe, “Deepfl-iqa: Weak supervision for deep iqa feature learning,”arXiv preprint arXiv:2001.08113, 2020

    Pith/arXiv arXiv 2001

  61. [61]

    Learning conditional knowledge distillation for degraded-reference image quality assessment,

    H. Zheng, H. Yang, J. Fu, Z.-J. Zha, and J. Luo, “Learning conditional knowledge distillation for degraded-reference image quality assessment,” inProc. of the IEEE/CVF Int. Conf. on Computer Vision (ICCV), 2021, pp. 10 242–10 251

    2021